dengyu-wu / spkeras Goto Github PK
View Code? Open in Web Editor NEWConversion from CNNs to SNNs using Tensorflow-Keras
License: MIT License
spkeras's People
Contributors
Stargazers
Watchers
spkeras's Issues
How the conversion process takes place ?
Hi, I have used SpKeras for a couple of applications, but still I am not sure of the conversion process from a CNN to SNN. I did refer this code, but wasn't able to get any satisfactory insights. So, can anyone help me on this ?
Bugs in SpKeras.2
Hello @Dengyu-Wu,
I've had the chance to try SpKeras.2 recently and have come across the below bugs.
- This one is fixed easily by commenting out the unseen file imports in the project.
spkeras/spkeras/layers/__init__.py", line 2, in <module>
from .MaxNorm_test import MaxNorm
ModuleNotFoundError: No module named 'networks.snn.spkeras.spkeras.layers.MaxNorm_test'
- I -tried to- fix this one via adding a control statement
if num+1 < len(lmax).
spkeras/spkeras/models.py", line 100, in convert
_bias = kappa*_weights[1]/lmax[num+1]
IndexError: list index out of range
- I -tried to- fix this one via adding a control statement as well
if num < len(l).
spkeras/spkeras/models.py", line 104, in convert
_weights[0] = kappa*_weights[0]/l[num]
IndexError: list index out of range
However, in the end another error gets thrown which I don't know how to get into. I tried running two different CNN models, one with BN layers, one without. Below error is thrown after trying the no-BN CNN:
spkeras/spkeras/models.py", line 179, in convert
layer.set_weights(weights[m])
File "/lib/python3.7/site-packages/tensorflow/python/keras/engine/base_layer.py", line 1783, in set_weights
'shape %s' % (ref_shape, weight.shape))
ValueError: Layer weight shape (65536, 1024) not compatible with provided weight shape (4096, 1024)
and this error belongs to the CNN with BN layers:
ValueError: Dimensions must be equal, but are 10 and 84 for '{{node spike_activation_6/Add}} = Add[T=DT_FLOAT](Placeholder, spike_activation_6/Mul)' with input shapes: [?,10], [84].
Any fixes?
getting pretty low accuracy with SNN
Hello @Dengyu-Wu,
I'm trying out spkeras on LeNet-5, yet the results are not satisfactory.
The org. ANN's accuracy is 0.9674, then it drops to 0.0980 when converted to SNN.
Can you please reason it out?
Below is the console output starting with evaluating the ANN's accuracy.
1875/1875 - 50s - loss: 0.2242 - accuracy: 0.9349 - val_loss: 0.0861 - val_accuracy: 0.9756
313/313 [==============================] - 5s 16ms/step - loss: 0.1085 - accuracy: 0.9674
||||| Model is ready. |||||
{'timesteps': 256, 'thresholding': 0.5, 'amp_factor': 100, 'signed_bit': 0, 'spike_ext': 0, 'epsilon': 0.001, 'use_bias': True, 'scaling_factor': 1, 'noneloss': False, 'method': 1}
Start Converting...
Extracting Lambda...
1/8AveragePooling2D
3/8AveragePooling2D
maximum activations: [1.0, 0.94426966, 0.99999297]
Building new model...
spikeforward_0_threshold: 100
spikeforward_1_threshold: 1.059012
spikeforward_2_threshold: 100
spikeforward_3_threshold: 1.059012
spikeforward_4_threshold: 100
spikeforward_5_threshold: 100
spikeforward_6_threshold: 100
New model generated!
||||| Model converted to SNN. |||||
Changing model timesteps...
New model generated!
{'timesteps': 256, 'thresholding': 0.5, 'amp_factor': 100, 'signed_bit': 0, 'spike_ext': 0, 'epsilon': 0.001, 'use_bias': True, 'scaling_factor': 1, 'noneloss': False, 'method': 1}
313/313 [==============================] - 6s 16ms/step - loss: nan - accuracy: 0.0980
||||| Model evaluated with accuracy 0.09799999743700027. |||||
I tried the code as in the usage example with different timesteps, threshold values, and scaling factors as well, but it didn't affect the output.
Note that I didn't use batch-normalization between the layers.
If the model created by spkeras can be exploited to neuromorphic hardware?
Hello @Dengyu-Wu,
I am evaluating the performance of spkeras 2.0 and trying to convert a CNN model to an SNN model for my publication.
However, I wonder if the model created by spkeras can be deployed to neuromorphic hardware directly such as SpiNNaker, TrueNorth or Loihi?
Joseph
AssertionError:
Hi,
I did add relu activation layer after each conv2d layer and the previous issue did get resolved. I am facing the following issue.
Issue:
---------------------------------------------------------------------------
AssertionError Traceback (most recent call last)
Cell In [13], line 7
2 from spkeras.spkeras.models import cnn_to_snn
4 #Current normalisation using cnn_to_snn
5 ##Default: signed_bit=0, amp_factor=100, method=1, epsilon = 0.001
----> 7 snn_model = cnn_to_snn(signed_bit=0)(cnn_model,x_train)
File /opt/ml_team_data/saurav/new/saurav/vgg16/SpKeras/spkeras/spkeras/models.py:29, in cnn_to_snn.__call__(self, mdl, x_train)
27 self.use_bias = use_bias
28 self.get_config()
---> 29 self.model = self.convert(mdl,x_train,
30 thresholding = self.thresholding,
31 scaling_factor = self.scaling_factor,
32 method = self.method,
33 timesteps=self.timesteps)
35 return self
File /opt/ml_team_data/saurav/new/saurav/vgg16/SpKeras/spkeras/spkeras/models.py:169, in cnn_to_snn.convert(self, mdl, x_train, thresholding, scaling_factor, method, timesteps)
167 new_mdl['config']['layers'] = layers
168 new_mdl = json.dumps(new_mdl)
--> 169 new_model = model_from_json(new_mdl,
170 custom_objects={'SpikeActivation':SpikeActivation})
171 input_shape = model.layers[0].input_shape
172 #new_model.build(input_shape)
173 #new_model = keras.Model(inputs=inputs, outputs=outputs)
File ~/.local/lib/python3.8/site-packages/keras/saving/model_config.py:109, in model_from_json(json_string, custom_objects)
86 """Parses a JSON model configuration string and returns a model instance.
87
88 Usage:
(...)
103 A Keras model instance (uncompiled).
104 """
105 from keras.layers import (
106 deserialize_from_json,
107 )
--> 109 return deserialize_from_json(json_string, custom_objects=custom_objects)
File ~/.local/lib/python3.8/site-packages/keras/layers/serialization.py:272, in deserialize_from_json(json_string, custom_objects)
266 populate_deserializable_objects()
267 config = json_utils.decode_and_deserialize(
268 json_string,
269 module_objects=LOCAL.ALL_OBJECTS,
270 custom_objects=custom_objects,
271 )
--> 272 return deserialize(config, custom_objects)
File ~/.local/lib/python3.8/site-packages/keras/layers/serialization.py:249, in deserialize(config, custom_objects)
212 """Instantiates a layer from a config dictionary.
213
214 Args:
(...)
246 ```
247 """
248 populate_deserializable_objects()
--> 249 return generic_utils.deserialize_keras_object(
250 config,
251 module_objects=LOCAL.ALL_OBJECTS,
252 custom_objects=custom_objects,
253 printable_module_name="layer",
254 )
File ~/.local/lib/python3.8/site-packages/keras/utils/generic_utils.py:734, in deserialize_keras_object(identifier, module_objects, custom_objects, printable_module_name)
731 custom_objects = custom_objects or {}
733 if "custom_objects" in arg_spec.args:
--> 734 deserialized_obj = cls.from_config(
735 cls_config,
736 custom_objects=dict(
737 list(_GLOBAL_CUSTOM_OBJECTS.items())
738 + list(_THREAD_LOCAL_CUSTOM_OBJECTS.__dict__.items())
739 + list(custom_objects.items())
740 ),
741 )
742 else:
743 with CustomObjectScope(custom_objects):
File ~/.local/lib/python3.8/site-packages/keras/engine/training.py:3034, in Model.from_config(cls, config, custom_objects)
3027 functional_model_keys = [
3028 "name",
3029 "layers",
3030 "input_layers",
3031 "output_layers",
3032 ]
3033 if all(key in config for key in functional_model_keys):
-> 3034 inputs, outputs, layers = functional.reconstruct_from_config(
3035 config, custom_objects
3036 )
3037 model = cls(
3038 inputs=inputs, outputs=outputs, name=config.get("name")
3039 )
3040 functional.connect_ancillary_layers(model, layers)
File ~/.local/lib/python3.8/site-packages/keras/engine/functional.py:1510, in reconstruct_from_config(config, custom_objects, created_layers)
1508 for layer_data in tf.nest.flatten(output_layers):
1509 layer_name, node_index, tensor_index = layer_data.as_list()
-> 1510 assert layer_name in created_layers
1511 layer = created_layers[layer_name]
1512 node_index = get_node_index(layer, node_index)
AssertionError:
My model architecture:
input_shape = (32, 32, 3)
input_layer = Input(input_shape)
layer = Conv2D(filters=4,
kernel_size=(1, 1),
strides=(1, 1),
padding="same")(input_layer)
layer=Activation('relu')(layer)
# 1
layer = Conv2D(filters=64,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(input_layer)
layer=Activation('relu')(layer)
# 2
layer = Conv2D(filters=64,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# Using Conv layer as a pooling layer
layer = Conv2D(filters=64,
kernel_size=(3, 3),
strides=(2, 2),
padding="same")(layer)
layer=Activation('relu')(layer)
# 3
layer = Conv2D(filters=128,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# 4
layer = Conv2D(filters=128,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# Using Conv layer as a pooling layer
layer = Conv2D(filters=128,
kernel_size=(3, 3),
strides=(2, 2),
padding="same")(layer)
layer=Activation('relu')(layer)
# 5
layer = Conv2D(filters=256,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# 6
layer = Conv2D(filters=256,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# 7
layer = Conv2D(filters=256,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# Using Conv layer as a pooling layer
layer = Conv2D(filters=256,
kernel_size=(3, 3),
strides=(2, 2),
padding="same")(layer)
layer=Activation('relu')(layer)
# 8
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# 9
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# 10
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# Using Conv layer as a pooling layer
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(2, 2),
padding="same")(layer)
layer=Activation('relu')(layer)
# 11
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# 12
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# 13
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# Using Conv layer as a pooling layer
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(2, 2),
padding="same")(layer)
layer=Activation('relu')(layer)
layer = Flatten()(layer)
layer = Dense(units=512)(layer)
layer=Activation('relu')(layer)
layer = Dense(units=10)(layer)
layer=Activation('relu')(layer)
model = Model(input_layer, layer)
My model summary:
Model: "model"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) [(None, 32, 32, 3)] 0
conv2d_1 (Conv2D) (None, 32, 32, 64) 1792
activation_1 (Activation) (None, 32, 32, 64) 0
conv2d_2 (Conv2D) (None, 32, 32, 64) 36928
activation_2 (Activation) (None, 32, 32, 64) 0
conv2d_3 (Conv2D) (None, 16, 16, 64) 36928
activation_3 (Activation) (None, 16, 16, 64) 0
conv2d_4 (Conv2D) (None, 16, 16, 128) 73856
activation_4 (Activation) (None, 16, 16, 128) 0
conv2d_5 (Conv2D) (None, 16, 16, 128) 147584
activation_5 (Activation) (None, 16, 16, 128) 0
conv2d_6 (Conv2D) (None, 8, 8, 128) 147584
activation_6 (Activation) (None, 8, 8, 128) 0
conv2d_7 (Conv2D) (None, 8, 8, 256) 295168
activation_7 (Activation) (None, 8, 8, 256) 0
conv2d_8 (Conv2D) (None, 8, 8, 256) 590080
activation_8 (Activation) (None, 8, 8, 256) 0
conv2d_9 (Conv2D) (None, 8, 8, 256) 590080
activation_9 (Activation) (None, 8, 8, 256) 0
conv2d_10 (Conv2D) (None, 4, 4, 256) 590080
activation_10 (Activation) (None, 4, 4, 256) 0
conv2d_11 (Conv2D) (None, 4, 4, 512) 1180160
activation_11 (Activation) (None, 4, 4, 512) 0
conv2d_12 (Conv2D) (None, 4, 4, 512) 2359808
activation_12 (Activation) (None, 4, 4, 512) 0
conv2d_13 (Conv2D) (None, 4, 4, 512) 2359808
activation_13 (Activation) (None, 4, 4, 512) 0
conv2d_14 (Conv2D) (None, 2, 2, 512) 2359808
activation_14 (Activation) (None, 2, 2, 512) 0
conv2d_15 (Conv2D) (None, 2, 2, 512) 2359808
activation_15 (Activation) (None, 2, 2, 512) 0
conv2d_16 (Conv2D) (None, 2, 2, 512) 2359808
activation_16 (Activation) (None, 2, 2, 512) 0
conv2d_17 (Conv2D) (None, 2, 2, 512) 2359808
activation_17 (Activation) (None, 2, 2, 512) 0
conv2d_18 (Conv2D) (None, 1, 1, 512) 2359808
activation_18 (Activation) (None, 1, 1, 512) 0
flatten (Flatten) (None, 512) 0
dense (Dense) (None, 512) 262656
activation_19 (Activation) (None, 512) 0
dense_1 (Dense) (None, 10) 5130
activation_20 (Activation) (None, 10) 0
=================================================================
Total params: 20,476,682
Trainable params: 20,476,682
Non-trainable params: 0
_________________________________________________________________
[Question] Does spkeras work on all Keras Conv layers?
Can you please let me know if spkeras works on all or only some of the conv layer types in the link @Dengyu-Wu?
I'm having a bit of a problem with a DepthwiseConv2D layer, hence am asking.
List index out of range
Hello!
I am again facing the same issue with the following architecture. Everything is same as before. Can you please help me on this?
input_shape = (32, 32, 3)
input_layer = Input(input_shape)
layer = Conv2D(filters=4,
kernel_size=(1, 1),
strides=(1, 1),
padding="same")(input_layer)
layer=Activation('relu')(layer)
# 1
layer = Conv2D(filters=64,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(input_layer)
# 2
layer = Conv2D(filters=64,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# Using Conv layer as a pooling layer
layer = Conv2D(filters=64,
kernel_size=(3, 3),
strides=(2, 2),
padding="same")(layer)
# 3
layer = Conv2D(filters=128,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# 4
layer = Conv2D(filters=128,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
# Using Conv layer as a pooling layer
layer = Conv2D(filters=128,
kernel_size=(3, 3),
strides=(2, 2),
padding="same")(layer)
layer=Activation('relu')(layer)
# 5
layer = Conv2D(filters=256,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
# 6
layer = Conv2D(filters=256,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# 7
layer = Conv2D(filters=256,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
# Using Conv layer as a pooling layer
layer = Conv2D(filters=256,
kernel_size=(3, 3),
strides=(2, 2),
padding="same")(layer)
layer=Activation('relu')(layer)
# 8
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
# 9
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# 10
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
# Using Conv layer as a pooling layer
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(2, 2),
padding="same")(layer)
layer=Activation('relu')(layer)
# 11
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
# 12
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
layer=Activation('relu')(layer)
# 13
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(layer)
# Using Conv layer as a pooling layer
layer = Conv2D(filters=512,
kernel_size=(3, 3),
strides=(2, 2),
padding="same")(layer)
layer=Activation('relu')(layer)
layer = Flatten()(layer)
layer = Dense(units=512)(layer)
layer = Dense(units=10)(layer)
layer=Activation('relu')(layer)
model = Model(input_layer, layer)
My model summary:
Model: "model"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) [(None, 32, 32, 3)] 0
conv2d_1 (Conv2D) (None, 32, 32, 64) 1792
conv2d_2 (Conv2D) (None, 32, 32, 64) 36928
activation_1 (Activation) (None, 32, 32, 64) 0
conv2d_3 (Conv2D) (None, 16, 16, 64) 36928
conv2d_4 (Conv2D) (None, 16, 16, 128) 73856
activation_2 (Activation) (None, 16, 16, 128) 0
conv2d_5 (Conv2D) (None, 16, 16, 128) 147584
conv2d_6 (Conv2D) (None, 8, 8, 128) 147584
activation_3 (Activation) (None, 8, 8, 128) 0
conv2d_7 (Conv2D) (None, 8, 8, 256) 295168
conv2d_8 (Conv2D) (None, 8, 8, 256) 590080
activation_4 (Activation) (None, 8, 8, 256) 0
conv2d_9 (Conv2D) (None, 8, 8, 256) 590080
conv2d_10 (Conv2D) (None, 4, 4, 256) 590080
activation_5 (Activation) (None, 4, 4, 256) 0
conv2d_11 (Conv2D) (None, 4, 4, 512) 1180160
conv2d_12 (Conv2D) (None, 4, 4, 512) 2359808
activation_6 (Activation) (None, 4, 4, 512) 0
conv2d_13 (Conv2D) (None, 4, 4, 512) 2359808
conv2d_14 (Conv2D) (None, 2, 2, 512) 2359808
activation_7 (Activation) (None, 2, 2, 512) 0
conv2d_15 (Conv2D) (None, 2, 2, 512) 2359808
conv2d_16 (Conv2D) (None, 2, 2, 512) 2359808
activation_8 (Activation) (None, 2, 2, 512) 0
conv2d_17 (Conv2D) (None, 2, 2, 512) 2359808
conv2d_18 (Conv2D) (None, 1, 1, 512) 2359808
activation_9 (Activation) (None, 1, 1, 512) 0
flatten (Flatten) (None, 512) 0
dense (Dense) (None, 512) 262656
dense_1 (Dense) (None, 10) 5130
activation_10 (Activation) (None, 10) 0
=================================================================
Total params: 20,476,682
Trainable params: 20,476,682
Non-trainable params: 0
_________________________________________________________________
The error:
---------------------------------------------------------------------------
IndexError Traceback (most recent call last)
Cell In [13], line 7
2 from spkeras.spkeras.models import cnn_to_snn
4 #Current normalisation using cnn_to_snn
5 ##Default: signed_bit=0, amp_factor=100, method=1, epsilon = 0.001
----> 7 snn_model = cnn_to_snn(signed_bit=0)(cnn_model,x_train)
File /opt/ml_team_data/saurav/new/saurav/vgg16/SpKeras/spkeras/spkeras/models.py:29, in cnn_to_snn.__call__(self, mdl, x_train)
27 self.use_bias = use_bias
28 self.get_config()
---> 29 self.model = self.convert(mdl,x_train,
30 thresholding = self.thresholding,
31 scaling_factor = self.scaling_factor,
32 method = self.method,
33 timesteps=self.timesteps)
35 return self
File /opt/ml_team_data/saurav/new/saurav/vgg16/SpKeras/spkeras/spkeras/models.py:101, in cnn_to_snn.convert(self, mdl, x_train, thresholding, scaling_factor, method, timesteps)
98 _weights[0] = _weights[0].astype(int)
99 _weights[0] = _weights[0]/2**bit
--> 101 _bias = kappa*_weights[1]/lmax[num+1]
102 _bias = _bias/norm
103 bias.append(_bias.tolist())
IndexError: list index out of range
IndexError: list index out of range
I have trained and saved a model and want to convert it to SNN by using SpKeras, but after running the code to convert I am facing the following issue:
---------------------------------------------------------------------------
IndexError Traceback (most recent call last)
Input In [10], in <cell line: 7>()
2 from spkeras.spkeras.models import cnn_to_snn
4 #Current normalisation using cnn_to_snn
5 ##Default: signed_bit=0, amp_factor=100, method=1, epsilon = 0.001
----> 7 snn_model = cnn_to_snn(signed_bit=0)(cnn_model,x_train)
File ~/Desktop/Toy network/spkeras/spkeras/models.py:29, in cnn_to_snn.__call__(self, mdl, x_train)
27 self.use_bias = use_bias
28 self.get_config()
---> 29 self.model = self.convert(mdl,x_train,
30 thresholding = self.thresholding,
31 scaling_factor = self.scaling_factor,
32 method = self.method,
33 timesteps=self.timesteps)
35 return self
File ~/Desktop/Toy network/spkeras/spkeras/models.py:101, in cnn_to_snn.convert(self, mdl, x_train, thresholding, scaling_factor, method, timesteps)
98 _weights[0] = _weights[0].astype(int)
99 _weights[0] = _weights[0]/2**bit
--> 101 _bias = kappa*_weights[1]/lmax[num+1]
102 _bias = _bias/norm
103 bias.append(_bias.tolist())
IndexError: list index out of range
My model:
Model: "model"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) [(None, 32, 32, 3)] 0
conv2d (Conv2D) (None, 32, 32, 4) 16
conv2d_1 (Conv2D) (None, 16, 16, 64) 2368
conv2d_2 (Conv2D) (None, 16, 16, 72) 41544
conv2d_3 (Conv2D) (None, 8, 8, 256) 166144
conv2d_4 (Conv2D) (None, 8, 8, 256) 65792
conv2d_5 (Conv2D) (None, 8, 8, 64) 16448
flatten (Flatten) (None, 4096) 0
dropout (Dropout) (None, 4096) 0
dense (Dense) (None, 100) 409700
dense_1 (Dense) (None, 10) 1010
=================================================================
Total params: 703,022
Trainable params: 703,022
Non-trainable params: 0
Any solutions ?
I also tried solving it by referring to this, but I wasn't able to.
Which neuromorphic hardware does SpKeras simulate ?
Hello,
I have been recently using SpKeras, but I am still not sure which neuromorphic hardware does it simulate ?
For example,
I used it to create an ANN, convert it to SNN and then simulate the SNN using SpKeras. But which neuromorphic hardware does SpKeras use to simulate SNNs ?
Also, the code the network architecture that I tried simulating is this
Low accuracy with bitwidth = 32
Hello @Dengyu-Wu ,
We are attempting to convert VGG16 network to snn, but are getting low accuracies whenever signed_bit is set to anything other than 0 (including 32, all other params unchanged). Please let us know if you've faced this issue before, or if it could be something on our end.
Output for Signed_bit = 0 :
{'timesteps': 256, 'thresholding': 0.5, 'amp_factor': 100, 'signed_bit': 0, 'spike_ext': 0, 'epsilon': 0.001, 'use_bias': True, 'scaling_factor': 1, 'noneloss': False, 'method': 1} 313/313 [==============================] - 35s 111ms/step - loss: 1624.8730 - accuracy: 0.7886
Output for Signed_bit = 32 :
{'timesteps': 256, 'thresholding': 0.5, 'amp_factor': 100, 'signed_bit': 32, 'spike_ext': 0, 'epsilon': 0.001, 'use_bias': True, 'scaling_factor': 1, 'noneloss': False, 'method': 1} 313/313 [==============================] - 39s 125ms/step - loss: 375521.1562 - accuracy: 0.1000
Thanks and regards,
Madhav
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.